Sunday, July 13, 2014

Rust for C++ programmers - part 8: destructuring

First an update on progress. You probably noticed this post took quite a while to come out. Fear not, I have not given up (yet). I have been busy with other things, and there is a section on match and borrowing which I found hard to write and it turns out I didn't understand very well. It is complicated and probably deserves a post of its own, so after all the waiting, the interesting bit is going to need more waiting. Sigh.

I've also been considering the motivation of these posts. I really didn't want to write another tutorial for Rust, I don't think that is a valuable use of my time when there are existing tutorials and a new guide in the works. I do think there is something to be said for targeting tutorials at programmers with different backgrounds. My first motivation for this series of posts was that a lot of energy in the tutorial was expended on things like pointers and the intuition of ownership which I understood well from C++, and I wanted a tutorial that concentrated on the things I didn't know. That is hopefully where this has been going, but it is a lot of work, and I haven't really got on to the interesting bits. So I would like to change the format a bit to be less like a tutorial and more like articles aimed at programmers who know Rust to some extent, but know C++ a lot better and would like to bring up their Rust skills to their C++ level. I hope that complements the existing tutorials better and is more interesting for readers. I still have some partially written posts in the old style so they will get mixed in a bit. Let me know what you think of the idea in the comments.

Destructuring


Last time we looked at Rust's data types. Once you have some data structure, you will want to get that data out. For structs, Rust has field access, just like C++. For tuples, tuple structs, and enums you must use destructuring (there are various convenience functions in the library, but they use destructuring internally). Destructuring of data structures doesn't happen in C++, but it might be familiar from languages such as Python or various functional languages. The idea is that just as you can create a data structure by filling out its fields with data from a bunch of local variables, you can fill out a bunch of local variables with data from a data structure. From this simple beginning, destructuring has become one of Rust's most powerful features. To put it another way, destructuring combines pattern matching with assignment into local variables.

Destructuring is done primarily through the let and match statements. The match statement is used when the structure being desctructured can have difference variants (such as an enum). A let expression pulls the variables out into the current scope, whereas match introduces a new scope. To compare:
fn foo(pair: (int, int)) {
    let (x, y) = pair;
    // we can now use x and y anywhere in foo

    match pair {
        (x, y) => {
            // x and y can only be used in this scope
        }
    }
}

The syntax for patterns (used after `let` and before `=>` in the above example) in both cases is (pretty much) the same. You can also use these patterns in argument position in function declarations:
fn foo((x, y): (int, int)) {
}

(Which is more useful for structs or tuple-structs than tuples).

Most initialisation expressions can appear in a destructuring pattern and they can be arbitrarily complex. That can include references and primitive literals as well as data structures. For example,
struct St {
    f1: int,
    f2: f32
}

enum En {
    Var1,
    Var2,
    Var3(int),
    Var4(int, St, int)
}

fn foo(x: &En) {
    match x {
        &Var1 => println!("first variant"),
        &Var3(5) => println!("third variant with number 5"),
        &Var3(x) => println!("third variant with number {} (not 5)", x),
        &Var4(3, St{ f1: 3, f2: x }, 45) => {
            println!("destructuring an embedded struct, found {} in f2", x)
        }
        &Var4(_, x, _) => {
            println!("Some other Var4 with {} in f1 and {} in f2", x.f1, x.f2)
        }
        _ => println!("other (Var2)")
    }
}

Note how we destructure through a reference by using `&` in the patterns and how we use a mix of literals (`5`, `3`, `St { ... }`), wildcards (`_`), and variables (`x`).

You can use `_` wherever a variable is expected if you want to ignore a single item in a pattern, so we could have used `&Var3(_)` if we didn't care about the integer. In the first `Var4` arm we destructure the embedded struct (a nested pattern) and in the second `Var4` arm we bind the whole struct to a variable. You can also use `..` to stand in for all fields of a tuple or struct. So if you wanted to do something for each enum variant but don't care about the content of the variants, you could write:
fn foo(x: En) {
    match x {
        Var1 => println!("first variant"),
        Var2 => println!("second variant"),
        Var3(..) => println!("third variant"),
        Var4(..) => println!("fourth variant")
    }
}


When destructuring structs, the fields don't need to be in order and you can use `..` to elide the remaining fields. E.g.,
struct Big {
    field1: int,
    field2: int,
    field3: int,
    field4: int,
    field5: int,
    field6: int,
    field7: int,
    field8: int,
    field9: int,
}

fn foo(b: Big) {
    let Big { field6: x, field3: y, ..} = b;
    println!("pulled out {} and {}", x, y);
}

As a shorthand with structs you can use just the field name which creates a local variable with that name. The let statement in the above example created two new local variables `x` and `y`. Alternatively, you could write
fn foo(b: Big) {
    let Big { field6, field3, ..} = b;
    println!("pulled out {} and {}", field3, field6);
}

Now we create local variables with the same names as the fields, in this case `field3` and `field6`.

There are a few more tricks to Rust's destructuring. Lets say you want a reference to a variable in a pattern. You can't use `&` because that matches a reference, rather than creates one (and thus has the effect of dereferencing the object). For example,
struct Foo {
    field: &'static int
}

fn foo(x: Foo) {
    let Foo { field: &y } = x;
}

Here, `y` has type `int` and is a copy of the field in `x`.

To create a reference to something in a pattern, you use the `ref` keyword. For example,
fn foo(b: Big) {
    let Big { field3: ref x, ref field6, ..} = b;
    println!("pulled out {} and {}", *x, *field6);
}

Here, `x` and `field6` both have type `&int` and are references to the fields in `b`.

One last trick when destructuring is that if you are detructuring a complex object, you might want to name intermediate objects as well as individual fields. Going back to an earlier example, we had the pattern `&Var4(3, St{ f1: 3, f2: x }, 45)`. In that pattern we named one field of the struct, but you might also want to name the whole struct object. You could write `&Var4(3, s, 45)` which would bind the struct object to `s`, but then you would have to use field access for the fields, or if you wanted to only match with a specific value in a field you would have to use a nested match. That is not fun. Rust lets you name parts of a pattern using `@` syntax. For example `&Var4(3, s @ St{ f1: 3, f2: x }, 45)` lets us name both a field (`x`, for `f2`) and the whole struct (`s`).

That just about covers your options with Rust pattern matching. There are a few features I haven't covered, such as matching vectors, but hopefully you know how to use `match` and `let` and have seen some of the powerful things you can do. Next time I'll cover some of the subtle interactions between match and borrowing which tripped me up a fair bit when learning Rust.

6 comments:

  1. Anonymous8:42 pm

    Good idea to target C++ programmers!

    ReplyDelete
  2. This is great, I am enjoying reading these posts, please carry on!

    Re: destructuring
    It seems very common in functional languages (e.g. Haskell) but not in C++ (you would just access the member variables of an object). Rust seems to favour the former, do you know why?

    ReplyDelete
  3. In other words:
    fn foo(b: Big) {
    let Big { field6, field3, ..} = b;
    println!("pulled out {} and {}", field3, field6);
    }

    ... you wouldn't write in C++:
    void foo(Big b) {
    int field6 = b.field6;
    int field3 = b.field3;
    cout << "pulled out " << field3 << " and " << field6 << endl;
    }

    ReplyDelete
  4. @Peter Hull: partly necessity, unlike C++, Rust has many data structures where you can't extract the data with field access - you have to destructure (enums, tuples, etc.). Partly just because destructuring is a somewhat modern construct and wasn't widely thought useful when C++ came about, but is now. Probably lots of other subtle, trade-off reasons too.

    ReplyDelete
  5. Thanks for your ideas. You can also find the details on Affity Solutions, at the C++ Development. The main object of the Affity Solutions is to provide quality web services and is among the few software development company in Nagpur.

    ReplyDelete
  6. Your blog is a heaven for those who want to learn C++ in Urdu

    ReplyDelete